The aim of the work is to build and realization variational algorithms for identifying the capacities of a series of pollution sources in a passive admixture transport model. The paper considers an example of variational assimilation of data on the concentration of suspended matter in the upper layer based on minimizing the prediction quality functional and an algorithm based on variational filtration of linear systems of equations. In both cases, solving adjoint problems is an essential element of the algorithms. The paper considers an example of variational assimilation of data on the concentration of suspended matter in the upper layer of the Sea of Azov in the area of the Long Spit, which is of great importance for navigation in this area.